As Qualcomm and Meta have previously collaborated,Llama 2 Large Language ModelMediaTek also announced earlier that it will deploy the Llama 2 large language model on its APU artificial intelligence computing components and artificial intelligence development platform NeuroPilot, emphasizing that its processor products can also handle artificial intelligence computing services that use large language models. It also announced that its upcoming flagship processor will be released at the end of the year.
MediaTek said that its flagship processor products will also be able to apply automatically generated artificial intelligence services and realize the use of Llama 2 large language models on terminal devices to respond to various possible artificial intelligence applications.
Compared to the past, large-scale language models are all processed through cloud-based collaborative computing. Users must first upload data to the cloud and wait for the results to be transmitted back to the device for use after processing is completed. The process may be delayed due to network transmission, and the data may also be at risk of privacy leakage due to being uploaded to the cloud. Therefore, in recent years, more and more industries have begun to promote device-side edge computing, allowing most calculations to be completed on the device side. Since data is only processed on the device side, there will be no privacy leakage issues, and the operation response speed can also be accelerated.
Even though many large language models that can be used on devices have simplified parameter sizes, they still require a certain level of computing power. Therefore, the device-side processor must have sufficient computing power to support it, and it must also have a certain degree of networking performance. It even requires an optimized computing architecture design for large language model operations. Only in this way can the benefits of running large language models on the device be realized.
Currently, both Qualcomm and MediaTek will announce their flagship processor products before the end of this year. Apple is also expected to use the new A-series processor in the iPhone 15 series. Samsung and Google also plan to launch new processor products, which are expected to have the ability to run large language models to meet the increasingly popular demand for automatic generative artificial intelligence applications.
